Power Flow-chart

[abishur]

I'm first going to preface this by telling myself to relax. The r-pi team has a lot more important things going on and they'll release the information when they have more free time (like after the board is actually released _

Okay that said, I was hoping we could get a flowchart look at the way power flows through the r-pi. We've had a lot of people make some very intelligent assumptions about the flow, in fact here's my basic understanding of the power flow.



Now obviously this is a high level look at the board, I didn't want to list every capacitor or resistor. It also ignores the SPI/DSI pins and, more importantly, it doesn't include RG1 which is located up by the GPIO pins. But as we're making assumptions as to how we can modify/power the board, I was wondering if we could get an official look at how power runs through the board.

[hippy]

I thought polarity protection and over-voltage protection had been removed because both are ensured by using a correctly wired micro-USB connection, but I do recall Gert saying reverse polarity protection remains. I'm not sure why, and I queried the voltage drop that would introduce, but if it's there it's there.

Other than that and the the additional regulators what you have generally matches with how I think things are ...

[liz]

OK - and I really *am* going to shut the computer down and get back on with being on holiday now, but seeing as I feel bad about not being around so much this week, here's a treat for you. This is the way the PSU works on the beta boards (with the obvious caveats about this not necessarily being a final version, and the potential for last minute changes, which I have to make for legal reasons). Click to enlarge.

[hippy]

Many thanks Liz.

[abishur]

wow, thanks liz! Now seriously go play on your vacation!

[nullstring]

Thanks Liz.
Unfortunately, it looks like we're burning all the excess power when dropping from 5v to any of the other voltages..

on the bright side, it looks like we could easily cut the legs of the 3v3 regulator and replace it with an outside source.

[Wooloomooloo]

Quote from nullstring on December 20, 2011, 07:36
Unfortunately, it looks like we're burning all the excess power when dropping from 5v to any of the other voltages..
Based on an older post (http://www.raspberrypi.org/for.....ostid-1936) the RasPi itself appears to eat about 0.4W of power, which if all drawn from the 1.6V rail gets you about 250mA, resulting in the horrifying amount of about 0.85W of wasted power on the linear regulators. Which is certainly non-negligible for battery-powered applications, but not even yawn-worthy in any other context, 'mkay?

[nullstring]

Quote from Wooloomooloo on December 20, 2011, 09:10
Quote from nullstring on December 20, 2011, 07:36
Unfortunately, it looks like we're burning all the excess power when dropping from 5v to any of the other voltages..
Based on an older post (http://www.raspberrypi.org/for.....ostid-1936) the RasPi itself appears to eat about 0.4W of power, which if all drawn from the 1.6V rail gets you about 250mA, resulting in the horrifying amount of about 0.85W of wasted power on the linear regulators. Which is certainly non-negligible for battery-powered applications, but not even yawn-worthy in any other context, 'mkay?

I don't disagree. (Although, I didn't check the math) Here's hoping that the 1v8 regulator doesn't sink my current.

The thing I am confused about is that Gert mentioned that there was a buck regulator for one of the lower voltage on chip(1v3 or 1v5 or something?).
While I am thankful to Liz for supplying the scematic, we still don't know what each rail is connected to.. or how much current each rail will use. Still plenty of unknowns.

[hippy]

Quote from nullstring on December 20, 2011, 16:37
The thing I am confused about is that Gert mentioned that there was a buck regulator for one of the lower voltage on chip(1v3 or 1v5 or something?).

That was my understanding. Presumably it was the intention to use that but, for whatever reason, got changed.

[Gert van Loo]

That was on the alpha boards. We switched to LDOs because they are cheaper. Not many people noticed but we have dropped the 'it uses only one watt' statement. It was dropping that or raising the price. The price was more important. I also wrote an article why an LDO burns more power then a buck regulator.

[bradburts]

I noticed
Time, power, cost - which one is better, only one way to find out...

I would be very interested in a 'dumb softees' summary on how the power works when its all figured & especially how I could get back to 1W.

Would you link the LDO article plz?
I partly followed the Pi on battery discussions but they keep getting a little too detailed for me so I need to learn.

[nullstring]

Quote from Gert van Loo on December 20, 2011, 16:51
That was on the alpha boards. We switched to LDOs because they are cheaper. Not many people noticed but we have dropped the 'it uses only one watt' statement. It was dropping that or raising the price. The price was more important. I also wrote an article why an LDO burns more power then a buck regulator.

It will take a few minutes to find a quote, but I could've sworn you said that in the beta boards, there was still one buck voltage regulator that was actually inside the SoC... Give me a minute.

[Bakul Shah]

Eben's Oct 20 blog entry is what confused people:
After a lot of experiments and spreadsheet work, we finally settled on an LDO-only power supply design, with a fixed 5V input, and the 1V2 core voltage generated directly from the input using the internal switch-mode supply on the BCM2835 die.


The linear regulator power loss will affect battery life if used as a portable device with a touchscreen etc but that doesn't seem to be the common use for its target audience. This is something us hobbyists have to be aware of. Chopping off inboard regulators may or may not work for experimenting.

[nullstring]

Ok, I don't know where my Gert quote went
This is actually a quote from Eben.
After a lot of experiments and spreadsheet work, we finally settled on an LDO-only power supply design, with a fixed 5V input, and the 1V2 core voltage generated directly from the input using the internal switch-mode supply on the BCM2835 die. We have chosen a 5V micro-USB jack to supply power to the board, for two reasons:

Then apparently we actually have the following voltages on the RPI: 5v, 3v3, 2v5, 1v8, 1v2.
With 1v2 possibly being internally to the chip only, and the only switch-mode regulation on the board.

[Wooloomooloo]

Quote from bradburts on December 20, 2011, 17:03
I would be very interested in a 'dumb softees' summary on how the power works when its all figured & especially how I could get back to 1W.

Would you link the LDO article plz?
I partly followed the Pi on battery discussions but they keep getting a little too detailed for me so I need to learn.
I think this is the article they are talking about:

Anyway, if one would really want to reduce power consumption the best way would probably be to a) try to use the lowest possible power mode of the SoC (playing Quake III is guaranteed to consume more than simply idling), and b) remove the 3V3 linear regulator and/or the 2V5 one from the board and replace it with some form of switching regulator(s) for the same voltage. Actually, it would probably make sense to figure out which voltage of the 3V3/2V5/1V8 is actually eating most of the current (one of them will definitely use the bulk of it), and generating that voltage with a switching regulator.

Even so, It should be noted that not every switching regulator is better than 90% efficiency at all (70-80% models still exist), and that efficiency is usually specified as "peak" efficiency wherever that would fall over the domain of power consumed (usually at nominal current), and at less than that current efficiency tends to go seriously down; so you should crunch the numbers before committing to this path.

[nullstring]

Quote from Wooloomooloo on December 20, 2011, 18:50
Quote from bradburts on December 20, 2011, 17:03
I would be very interested in a 'dumb softees' summary on how the power works when its all figured & especially how I could get back to 1W.

Would you link the LDO article plz?
I partly followed the Pi on battery discussions but they keep getting a little too detailed for me so I need to learn.
I think this is the article they are talking about:

Anyway, if one would really want to reduce power consumption the best way would probably be to a) try to use the lowest possible power mode of the SoC (playing Quake III is guaranteed to consume more than simply idling), and b) remove the 3V3 linear regulator and/or the 2V5 one from the board and replace it with some form of switching regulator(s) for the same voltage. Actually, it would probably make sense to figure out which voltage of the 3V3/2V5/1V8 is actually eating most of the current (one of them will definitely use the bulk of it), and generating that voltage with a switching regulator.

Even so, It should be noted that not every switching regulator is better than 90% efficiency at all (70-80% models still exist), and that efficiency is usually specified as "peak" efficiency wherever that would fall over the domain of power consumed (usually at nominal current), and at less than that current efficiency tends to go seriously down; so you should crunch the numbers before committing to this path.

Since the 3V3 rail is brought out to the GPIO pins, it should be trivial to replace it assuming that cutting traces isn't that hard. (Something I've never done before).

The 2v5/1v8 regulators would be much harder to replace and it's unlikely it's worth the effort in doing so.

[Gert van Loo]

I don't know what happened but my reply from an hour ago is not here!
Again:
I admit there is a SMPS inside the BCM2835 which supplied the 1V2. I did not want to make things more complex then required.
You normally start with making the highest voltage switched, as it supplies the current to the lower voltages as well. Also : the 5V is used for HDMI and USB and the BCM2835 SMPS , if you have an application where you do not need HDMI/USB (e.g. Ethernet only) you do not need the 5V as the SMPS should run of 3V too (never tried it though!!) . So you could start with a 3V supply (battery??) that will save a lot of power. Don't bother with optimizing the 2v5 it is a very low output current.

[nullstring]

Quote from Gert van Loo on December 20, 2011, 21:05
I don't know what happened but my reply from an hour ago is not here!
Again:
I admit there is a SMPS inside the BCM2835 which supplied the 1V2. I did not want to make things more complex then required.
You normally start with making the highest voltage switched, as it supplies the current to the lower voltages as well. Also : the 5V is used for HDMI and USB and the BCM2835 SMPS , if you have an application where you do not need HDMI/USB (e.g. Ethernet only) you do not need the 5V as the SMPS should run of 3V too (never tried it though!!) . So you could start with a 3V supply (battery??) that will save a lot of power. Don't bother with optimizing the 2v5 it is a very low output current.



You might not know this, but does MIPI DSI require 5v?
If not, this might be a good reason to target MIPI DSI for any portable applications with a display.

Also will USB still work with only 3v3? Obviously the V+ line on the USB port will not receive the adequate voltage, but if you are only interested in the data lines, which IIRC are 3v3, will it still work?

[dp11]

Just a few ideas :

1) It might be nice to meet the usb power specs which limit devices to 10uF. The circuits have 220uF currently. I know only power is being used but it might be nice to be a little more friendlier.

2) In my experience 1nF (C2) as decoupling rarely makes any difference. Usually via placement is more important. Unless anything on the circuits actually require 1nF just replacing them all with 10nF or 100nF will typically lower the noise floor and reduce bom costs very slightly.

If you need any help please let me know. I can sign NDAs if required.

[Gert van Loo]

No DSI/CSI do not require 5V. The MIPI PHY specification has an upper limit of 600mV. (That is signaling level + common mode voltage). I guess you can make a phy run from as little as 1.2 V.

As to USB and no 5V: I said that you can probably run of 3.3V Ifyou do not use USB/Ethernet.

[arm2]

Since the forum change over The diagrams doesn't seem to be able to be viewed large enough to read:-(  [Using FireFox 3.6.25]

[TheTap]

OK here   [Using FireFox 9.0-0.2.1]

[hippy]

arm2 said:


Since the forum change over The diagrams doesn't seem to be able to be viewed large enough to read:-(  [Using FireFox 3.6.25]


Same problem with FF 5.0 – They enlarge but not to the full / native size they are. Also same on Chrome 16.0.912.63 m. Seems to be an issue of limiting to browser / screen height.

[slacker]

Same here (Seamonkey 2.6), right clicking on the thumbnail and doing "Open link in new window" opens it full size though.

[arm2]

Yes right clicking over the image, selecting 'open in new window' and then in the new window right? clicking, fully enlarges it, to a readable size!